Document analysis systems

在我们这个信息丰富的社会中，文件被广泛使用。 拟议的研究简化了纸质文档和电子文档之间的过渡，有助于灵活和通用的技术，将扫描的纸质文档中的信息转换为电子形式。 一个拟议的项目开发了新的方法来表示由文档识别系统做出的决定。 决策的显式表示使得确定识别错误的原因变得容易，并且支持识别算法的自动优化。 第二个拟议项目通过将文件文本分析与文件数字分析相结合，改进文件处理。 该方法将在一个文件分类问题上进行测试：确定哪些技术文件与维护生物医学数据库有关。 第三个拟议项目开发使用外部数据验证和纠正识别结果的方法。 例如，互联网搜索可用于验证从扫描的名片识别的信息，并纠正电话号码中识别错误的数字。

Documents are widely used in our information-rich society. The proposed research streamlines the transition between paper documents and electronic documents, contributing flexible and general techniques for converting the information in scanned paper documents into electronic form. One proposed project develops new methods for representing the decisions that are made by document recognition systems. An explicit representation of decisions makes it easy to determine the causes of recognition errors, and supports automatic optimization of recognition algorithms. The second proposed project improves document processing by combining an analysis of document text with an analysis of document figures. The approach will be tested on a document classification problem: identify which technical documents are relevant for maintaining a biomedical database. The third proposed project develops methods of using external data to validate and correct recognition results. For example, an internet search can be used to verify information recognized from a scanned business card, and to correct incorrectly-recognized digits in a telephone number.